Relevance Search via Bipolar Label Diffusion on Bipartite Graphs
The task of relevance search is to find relevant items to some given queries, which can be viewed either as an information retrieval problem or as a semi-supervised learning problem. In order to combine both of their advantages, we develop a new relevance search method using label diffusion on bipartite graphs. And we propose a heat diffusion-based algorithm, namely bipartite label diffusion (BLD). Our method yields encouraging experimental results on a number of relevance search problems.
